ioi said:
Well, it depends on how deeply you wish to get into all of this! Past trends are always informative and helpful, although of course not directly applicable. You have to think of a sales estimate as a combination of things - samples of raw data from retailers, user data that we can monitor, the application of past trends and a little bit of fitting / fudging to allow for systematic errors. I'll try to give some examples. We may receive some sell-through data from a chain of retailers - say 20-30 stores in a particular region. Assuming that this region is representative of the country as a whole (or if not, applying some factors to try to adjust for known systematic differences) you can extrapolate up to estimate how many units that retailer sold countrywide by knowing how many stores there are in total. So if the raw data says that 20 stores sold 275 Xbox One units and we know that it is representative of the whole USA and there are 7000 stores in total then we can estimate that this chain sold 275 / 20 * 7000 = 96k units. Clearly, if the 20 stores we have data for are bigger than the national average store size for this chain then that would inflate our estimate for this retail chain (for all platforms) vs reality. Similarly, if there is something about this region that may lend itself to more people buying an Xbox One than other regions in the country then it would skew the data for this platform upwards relative to others. Or if another major retailer in this region had a better deal for this platform that wasn't reflected countrwide then that would make our sample under-representative and the data would be low. Hopefully you can see the limitations and complexity here and how you have to make a number of assumptions - each one of which will harm accuracy. You combat this by repeating the process for 5-10 different samples and taking a weighted average to hopefully start to arrive at a more accurate figure. One key method in trying to increase accuracy is to group samples into retail groups. So we know that Walmart and Target generally show similar sales patterns (general stores with video game sections), Frys and Best Buy are general electronics retailers and so on. We also split online and in-store as online tends to skew more towards intended purchases while in-store will have more impulse buys. Understanding how the market is split and how big each demographic is can help us to extrapolate data in a representative way. Using a crude example - if we estimate that Walmart sold 100k Wii U units via hard data and Target had the exact same deals on but we don't have any hard data from them then our best estimate of Target's sales would be to use Walmarts and adjust by the relative marketshares - so it Walmart is 30% and Target is 9% then we'd know that around 30k WiiUs were sold at Target which should be a fairly good estimate as they are similar types of store with similar demographics. This would give a total of 130k WiiUs in our "general retailers group" which has an overall marketshare of ~40%. If we based the data purely on this, it would suggest something like 325k in total but of course the general retail group (aside from toy stores) generally favours Nintendo products much more than electronics retailers or gaming retailers like Gamestop. Our estimate for Gamestop (35% marketshare) might only be 50k units. This is why you need to take a weighted average and we do that by retail groups - you'd get very different data if you used only the general retail data (130k units from ~40% of the overall market) or only the gaming retail data (50k units from 35% of the market). Finally, past trends. There's no reason why Xbox One and PS4 shouldn't follow broadly similar trends to Xbox 360 and PS3. So if we typically see a 3x lift on Black Friday week vs an average of the 4 weeks before then the same should be true in this case. This provides another estimate to throw into the mix or at the very least a sanity check. Some products always see bigger holiday lifts than others - Xbox a little more than Playstation, Nintendo more than everyone else, kids games more than adult-themed games and so on - patterns are clear to see when you examine data as much as we do and you have to sometimes make assumptions that those same patterns will continue to hold this year. All-in-all, estimating sales is a complex process and we always do the best we can with the data we have. |
Thanks for the insight, it sounds like simpler process than I expected it to be, but at the same time I can now better understand why there are quite big (in some cases) differences between VGC and official figures. You've undertaken a mammoth task and are doing it well, keep going!
